Don Norman, Forgive Us Machine-Centered Design Is Here (medium.com)

AI agents are no longer just interfaces we use — they’re becoming first-class “users” of the web. Authors Edi Bianco and John Petitte argue that as assistants like ChatGPT, Gemini and AI browsers (Comet, Atlas) read, aggregate and act on our behalf, the old human-centered UX model is due for a rethink. Instead of colorful layouts and micro-interactions designed to capture human attention, the web may evolve into a semantic, machine-first layer where intent is expressed to an agent, actions are executed without human-facing pages, and emotional “delight” shifts to the AI layer. That flip matters to designers, product teams and the AI/ML community because it changes who the primary consumer of content is, how trust is established, and where effort should be invested: human affordances vs. machine-readable signals. Technically, this implies new primitives: richer metadata and semantic markup (an “AI version” of sites), machine-to-machine authentication and delegated identity tokens, and payment/consent protocols that let an agent prove authority without human-driven forms. It also magnifies existing risks—confidence can masquerade as truth, invisible consent raises ethical and regulatory questions, and centralized control of machine channels could enable silent manipulation. Practitioners should prepare by building verifiable provenance, contextual grounding and ethics into models and protocols, designing dual-layer experiences (emotional for humans, structured for agents), and helping standards evolve for secure, auditable agent-to-service interactions.